Topics: What Do You Do When There is No Model?
on General Discussion
What Do You Do When There is No Model?
In 30+ years of building power supplies, I never met one yet that didn't have a surprise in store for me.
So what do you do when you run into something that doesn't behave as expected?
09-09-2013 11:11 PM
Let me elaborate on the question. There is a wide spectrum of design approaches in the industry, ranging from 100% hardware focused, to heavy software focus.
Option 1: Some engineers have been building power supplies for years. They do no modeling, no simulation, they just build what they have always built, test it thoroughly, and ship it.
Option 2: Some software may be used to guide the design better, but when the two don't match, the software is abandoned and the hardware measurements rule the day.
Option 3: Software is used extensively, and a lot of work is put into refining the models to get the two to match. But there will still be areas where hardware deviations will exist, and practicality dictates the product must be shipped with just empirical results in some areas.
Option 4: The software rules. If the hardware cannot be 100% predicted, a different architecture or topology must be used to provide a complete match.
So which category do you go for?
Which will give the most rugged design?
09-09-2013 11:12 PM
If it works - don't break it. So Option 1 would seem to be a great method - but of course it requires lots of experience - which comes with, either a lot of luck, or most likely a fair share of failures - or challenges that had to be overcome. And just because something hasn't jumped out to bite - doesn't mean it isn't there lurking waiting to pounce.
Option 2: sometimes you can spend too much time with the software and not enough with the hardware. Before abandoning the software - it is great to know if it is truly the software that is the problem - or whether it is a problematic circuit that is operating on the fence. Each engineer has his/her feeling about their competency in each area and has to make that choice which to trust.
Option 3 requires experience as well - I think you have to know what to expect or have experience to know what parameters to model - or you get only a simple simulation that may not be worth much when you go to the lab. Extensive use of behavior models are great because they speed up the simulation - but what real information do they provide other than show a simple output that was expected.
Option 4; many engineers actually practice this - they choose a topology because a vendor offers it in a simulation - even if it is only a behavior model. It is too bad. I suppose the chosen topology might work great, but I'd feel more confident in that engineer if the choice was not made for that reason.
For myself, I like to run simulations before I build, so I can play with things and watch it work as I intended. Sometimes I find things that I didn't intend, and other times I use it to optimize competing parameters to work out an operation that I am happy with. When I do go into the lab, I typically see what I expected. If I don't, then I can modify my model - and then I usually do see what I did in the lab - which gives me confidence that I have a good understanding of what it is and that it is under control.
Good simulations take time and experience - just like designing power supplies. I think it has helped my understanding and has replaced countless hours in the lab.
09-09-2013 11:12 PM
In my very few years of building dc/dc converters, I just used hand calculations or calculations that I put into Mathcad for inductor current ripple. I concentrated on getting what I thought was a good layout and then worked at the bench quite awhile.
09-09-2013 11:13 PM
Simulation models are only as good as the information that you put in them. I find that its best to mathematical design the converter go away and simulate the converter based on your calcualtions to ensure that it looks about right and then go and test. It can be very time consuming to input all of the data needed to make a simulation 'accurate' that its best to test it in real life as there is always something that you have left outside of your model that will trip you up.....
09-09-2013 11:13 PM
Option 1: That's how I've started: followed the advice of the older peers.
Option 2: Later on, the funniest was when I've stumbled upon the ZCS behavior of the forward DCDC having x_former with too much leakage inductance. Well it did not behave as THE model. Wound better x_former and it's gone. Few years later discovered F.Lee articles on the matter.
Option 3: Well sometimes it's better to make prototype. I would love to use hardware in the loop every time. The problem is usually the adequacy of the model especially fringing effects. At power level of a few megawatts experience is the must and too much optimism is too costly.
Option 4: have not gotten there yet...
09-09-2013 11:14 PM
Brian makes a good point with his Mathcad example that Spice is not the one-and-only software tool we use to analyze power converters. I've analyzed the power dissipation of complex motor drives in Excel for example, or whole subsystems of a satellite power converter in Mathcad. New inventions require the designer to create new tools. "What do you do when there is no model?" You create a new model or equation.
09-09-2013 11:14 PM
My very first power supply simulation was in 1980 for a ferro-resonant power supply. I used a TI programmable calculator to plot the state-plane diagrams to understand the nonlinear operation with a saturable magnetic. This was my first try at writing my own simulator to make it do exactly what I need.
Spice back then (if it existed?) was so hideously slow, that it would have been impossible to use.
What do you do when the modeling attempts don't match reality? When the top minds give you an equation that doesn't actually work very well because of nonlinearities or other strange events in the converter that defy prediction?
09-09-2013 11:15 PM
Ray, "...when the modeling attempts don't match reality? When the top minds give you an equation that doesn't actually work very well because of nonlinearities or other...". I bet many have seen the drawbacks of blindly using a model without studying it first. Usually I find that the model is a simplification of a circuit which doesn't quite match the new circuit being invented. Several years ago I used your model (we called it the 'current mode Ridley model') to analyze a PWM current source driving a magnetic structure for an ion accelerator. The first attempt (blindly plugging in the model) didn't work. To find why, I had to study your model for hours until it was clear. I was incorrectly using an average load voltage in the model. Because this circuit was a current source into a complex L-C load with a widely varying voltage, I found I had to change several fixed terms into variables. The final model matched the o'scope screen pretty well.
Thanks to you for introducing me to this elegant model (picture me holding up a beer) which incorporates the slope compensation into the feedback loop. It's great for standard DC/DC converters and has become the basis for many non-standard power converters.
09-09-2013 11:16 PM
I'm leaning toward option 2. The software provides a non-destructive means for testing various configurations and sometimes reveals design defects that could take a long time to find in hardware. However, the software is no better than the model and parasitic elements or field coupling are next to impossible to model. For example, I had a buck converter that was very stable. Then on the next build it oscillated. After a thorough investigation, the only thing we could find is that the choke was wound in the opposite direction. As it turns out, there was enough external field from the choke to induce a small voltage in the sense circuits. When the winding was reversed so was the phase of this signal which introduced a small but sufficient amount of positive feedback. On the other hand, I had another buck converter design that had good phase margin and stable dynamic response but the simulation showed that, if the input voltage was gradually raised from below the point of regulation into the regulating region, a non-linear oscillation would occur. It didn't show in the hardware because the source voltage had sufficient ripple that it masked the oscillation. When the regulator was powered from a lab supply, it oscillated.
09-09-2013 11:16 PM
I agree with Bruce and I liked his examples. They show that for some “details”, simulations are useless. No one would predict from simulation that the direction of the winding of the choke would influence the operation of the circuit. To see these effects, the only way is to build and test the power supply. And then change the project according to what was learned in the lab.
But for me the simulation is useful to get the “big picture” of how the power supply is going to work (especially the control circuits). Usually it is fast to create a simple model of the converter, with ideal switches, no (or very little) parasitic components etc. And then use this simple simulation to test the stability of the converter for different loads and so on.
Hopefully, the stability of the controller is usually not affected by small variations (the moment it does, you’re… well, lost) and the fast model can tell you whether the controller has good chances to work properly or not.
For everything that depends on parameter variations, parasitic components, noise, EMI, it is usually very hard to simulate and better to verify directly in the prototype. Especially EMI, that still looks like black magic to me.
Of course, the moment you have designed the same converter over and over again, only with small changes in the specs, you may ignore simulation and choose option 1. As this is not the case for me, simulation helps a lot.
09-09-2013 11:17 PM
To see whether your concept is going to work or not simulation is going to give your early warning and watch out point.. For that reason it is useful.. I guess they are complementary and should be used whenever is needed.. Big question is when is needed . Testing will not be replaced by simulation anyway.. So this is a must.. Then what to test is next question.. Test everything is not the answer.. It may take months to do that.. In this competitive world most of the time we do not have that luxury. Endless time to test. That could be possible in academic environment. But in real world you are always against dead line.. So as design engineer I look to the application and deeply analyze end user system. From this come up test methodology and parameter to be test in detail based on application.
Compare test results with the model result to convince myself that I did not miss " important " points in the design.
Another important point is PCB layout. the same circuitry my work as accepted if PCB layout is done " properly". may not work at all if it is not done properly..
So many variable when it comes to design, All available tools, experience, Matcad, simulation has its place in the design providing that used properly. Choosing one over the other may lead sens of wrong confidence at best or disaster...